Edge AI based Sparse Point Cloud Perception (ARD/302)

The advancement of 3D sensing technology has revolutionized the way industries detect objects of interest, with applications spanning across smart transport, smart logistics, smart factories, security, and surveillance. However, the technology's efficacy is often compromised in scenarios involving high-speed movement, severe weather, and other environmental factors, leading to the missed capturing of partial point clouds. Additionally, multiscale object detection remains a challenging task for 3D sensing technology. To overcome these challenges, the proposed project seeks to develop a sparse point cloud specific AI, with a focus on transportation, logistics and manufacturing applications. The project will enhance the technology's object detection capabilities in sparse and multiscale data scenarios. The project will entail developing three main functions: (1) a 3D object detection module, which will provide accurate and multiscale perception on 3D point clouds; (2) a 2D/3D feature fusion module, which will combine 2D image features and 3D graph features to improve accuracy and enable the capturing of small objects; and (3) an edge-based optimization module, which will enable the AI to embed into edge devices and perform onsite operations. The proposed edge-based sparse point cloud AI technology aims to provide companies with fast and precise point cloud perception results on sparse and multiscale tasks, resulting in efficient and better business outcomes. The technology's vast applications across various industries make it a game-changer, driving innovation in the field of artificial intelligence, and streamlining company operations.